Nvidia launches open-weight Nemotron models for enterprise agents and local AI PCs
The brief
Nvidia has introduced the 120-billion-parameter Nemotron 3.5 Super VL for enterprise agents, along with smaller Nemotron 3.5 Bolt models designed for local deployment on PCs and workstations using Nvidia chips. Super VL and Bolt 74B-A9B are available now, while the conversational Bolt E4B is expected later in October; Nvidia’s performance comparisons and customer-use claims have not been independently verified in the article.
Key points
- Nemotron 3.5 Super VL is a 120-billion-parameter open-weight model for enterprise agent tasks and is intended to offer accuracy comparable to the larger Nemotron 3 Ultra.
- Nemotron 3.5 Bolt 74B-A9B 24GB targets local agentic tasks such as multistep reasoning, tool calling, and web browsing on Nvidia RTX Spark PCs and DGX Spark workstations.
- The smaller Bolt E4B is designed for conversational agents and is scheduled to arrive later in October.
- Nvidia says Super VL is faster than named Anthropic and OpenAI models on selected post-trained agentic benchmarks, and lists Skydance, EY, Ramp, CrowdStrike, Perplexity, and Salesforce as users.
- The models support organizations running and post-training agents on their own infrastructure, while local PC deployment is positioned as a way to keep workloads off the cloud.
Read the source report
TechCrunch AI ↗Full summary
Nvidia announced the open-weight Nemotron 3.5 Super VL, a 120-billion-parameter model aimed at enterprise agents. The company says it uses fewer parameters than Nemotron 3 Ultra while delivering similar accuracy on instruction-following and tool-use tasks, including document intelligence, accounting, scientific experimentation, medical claims, and security resilience.
Nvidia also introduced two smaller Bolt models for local use on PCs with the Nvidia RTX Spark chip and workstations running DGX Spark. Nemotron 3.5 Bolt 74B-A9B 24GB is intended for multistep reasoning, tool calling, and web browsing, while the smaller Bolt E4B focuses on conversational agents.
Super VL and Bolt 74B-A9B are available now, according to Nvidia, and Bolt E4B is expected later this month. Nvidia says the systems can let organizations post-train models on their own data and deploy them on their own infrastructure; local deployment is also presented as a way for PC users to run agent workloads without sending them to the cloud.
Nvidia claims Super VL outperforms competing models from Anthropic and OpenAI on selected post-trained agentic tasks and says companies including Skydance, EY, Ramp, CrowdStrike, Perplexity, and Salesforce are using it. The announcement comes as enterprise customers explore open models and as AI developers consider alternatives to Nvidia hardware, including AWS Trainium, Cerebras, and internally developed chips; the article does not independently verify Nvidia’s benchmark or customer claims.